fix(channels): propagate audience on Slack history-fetched ChannelInput

[Aaronontheweb]

Summary

SlackThreadHistoryFetcher.ConvertMessageAsync built ChannelInput without setting the resolved historical audience. After PR #984 / #45f4c57b made hydration run once per actor lifetime through a Hydrating gate, fresh DM sessions get their backfill from history (the live inbound is dropped as stale by _pendingCursorTs advancing inside hydration). With Audience null on the produced ChannelInput, MessageSourceFactory.Create fell back to SessionPipelineOptions.DefaultAudience = TrustAudience.Public, silently denying shell_execute with tool_not_allowed_for_audience_profile — even when the operator had Slack.ChannelAudiences["dm"] = "personal" configured.

Discord's twin (DiscordThreadHistoryFetcher) already sets Audience correctly. Slack regressed at the original introduction (#576) and the recent hydration refactor exposed it on the very first DM message of a fresh session.

Symptom observed in session D0AC6CKBK5K/1778736039.290249: a Personal-audience DM saw shell tools blocked.

Fix

One-line addition in SlackThreadHistoryFetcher.ConvertMessageAsync:

return new ChannelInput
{
    SenderId = senderId,
    ChannelId = channelId.Value,
    MessageId = $"{channelId.Value}:{message.Ts ?? string.Empty}",
    Audience = audience,   // ← was missing
    Contents = contents,
    ReceivedAt = receivedAt
};

The audience value is already resolved upstream in FetchRepliesAsync via ResolveHistoricalAudience and passed down — it just wasn't propagated to the resulting ChannelInput.

Tests

• SlackThreadHistoryFetcherTests.Hydrated_channel_input_carries_resolved_historical_audience — regression test that fails on the pre-fix code.
• DiscordThreadHistoryFetcherTests.Hydrated_channel_input_carries_resolved_historical_audience — lock-in test for the already-correct Discord twin so it can't regress symmetrically.

Follow-up

Filed as a separate issue: make ChannelInput.Audience a required property so future callsites can't omit it and silently fall back. That refactor touches every new ChannelInput { ... } callsite plus a handful of tests; out of scope for this hotfix.

Test plan

• Targeted xUnit runs of both new tests pass.
• dotnet slopwatch analyze clean (0 issues).
• ./scripts/Add-FileHeaders.ps1 -Verify clean.